GPI: GenAI-Powered Inference Logo

Getting Started

  • Installation
    • Requirements
    • From PyPI (Version 0.2.1)
    • From Source
  • What Is GPI?
    • GPI Workflow
    • Current Package Scope
  • How to Use a GPU
    • Checking CUDA
    • Using a Hosted GPU
    • Using Remote Model Execution
  • Questions and Bug Reports

Data Generation

  • Generating Texts with Llama 3
    • Loading Llama 3.1
    • Creating Texts
    • Repeating Existing Texts
    • Pooling
    • Custom System Instruction
    • Additional Arguments
  • Generating Texts with Other LLMs
    • Loading Gemma 2
    • Generating and Saving Representations
    • Mean Pooling Across Generation Steps
  • Generating Texts without a Local GPU
    • Installing and Authenticating
    • Choosing an Available Model
    • Remote Generation and Hidden States
    • Important Notes
  • Generating Images with Stable Diffusion
    • How the Image Representation Is Constructed
    • Using extract_images
    • Transforming Multiple Images
    • Arguments
  • Generating Videos using Cosmos Tokenizer
    • What is Cosmos Tokenizer?
    • Installing the Video Dependencies
    • How to use Cosmos Tokenizer
    • Arguments
    • Output Files
    • Command-Line Interface
    • Important Notes
  • Transcribing Audio using Whisper
    • What is Whisper?
    • Installing Whisper
    • Choosing a Whisper Model
    • Command-Line Usage
    • Python Usage
    • Transcribing Audio from a Video
    • Aligning Transcripts with Video Segments
    • Important Notes

Basic Operations

  • Text-As-Treatment
    • What is Text-As-Treatment?
    • How to estimate treatment effects
      • Step 1: Load the Internal Representations
      • Step 2: Estimate the Treatment Effects
    • How to control confounders
      • Method 1: Using a Formula with a DataFrame
      • Method 2: Using a Design Matrix
    • Visualizing Propensity Scores
    • Hyperparameters
  • Text-As-Confounder
    • What is Text-As-Confounder?
    • How to estimate treatment effects
  • Image-As-Treatment
    • How to estimate treatment effects
      • Step 1: Load the Internal Representations
      • Using Flattened Representations
      • Using a Convolutional Neural Network
      • Step 2: Estimate the Treatment Effects
  • Video-as-Treatment
    • Overview of the Data
    • Step 1: Extract the Video Representations
    • Step 2: Align and Pad the Sequences
    • Step 3: Estimate the Intervention Curve
    • Real-Time Outcomes
    • Segment-Specific Interventions
    • Modality and Audio Support

Advanced Operations

  • Hyperparameter Tuning
    • Automated Hyperparameter Tuning
    • Static Hyperparameter Tuning
      • Static Search Space
    • Dynamic Hyperparameter Tuning
      • Dynamic Search Space
      • Repeated Outcomes
    • Important Notes
  • Customizing Your Analysis
    • TarNet
    • Propensity Score Model
    • Cross-Fitted Score Contributions
    • Complete Cross-Fitting
  • When an LLM Is Too Large
    • Installing bitsandbytes
    • Loading a 4-bit Model

References

  • Function Reference
    • Text Generation
    • Image Generation
    • Video Processing
    • Static Inference
    • Dynamic Inference
  • dml_score
    • Description
    • Arguments
    • Returns
    • Example Usage
  • estimate_k_ate
    • Description
    • Arguments
    • Returns
    • Example Usage
  • estimate_psi_split
    • Description
    • Arguments
    • Returns
    • Example Usage
  • extract_and_save_hidden_states
    • Description
    • Arguments
    • Returns
    • Example Usage
  • generate_text
    • Description
    • Arguments
    • Pooling Behavior
    • Returns
    • Example Usage
  • get_instruction
    • Description
    • Arguments
    • Returns
    • Example Usage
  • load_hiddens
    • Description
    • Arguments
    • Returns
    • Example Usage
  • save_generated_texts
    • Description
    • Arguments
    • Returns
    • Example Usage
  • pad_to_multiple_of_8
    • Description
    • Arguments
    • Returns
    • Example Usage
  • StableDiffusionImg2ImgExtractor
    • Description
    • Parameters
    • Example Usage
    • Methods
      • preprocess_image
      • encode_prompt
      • get_hidden_states
      • decode_latents
      • transform_image
      • save_image
  • extract_images
    • Description
    • Arguments
    • Returns
    • Example Usage
  • CosmosVideoExtractor
    • Description
    • Parameters
    • Example Usage
    • Methods
      • preprocess_video
      • encode_video
      • decode_latents
      • reconstruct_video
      • transform_video
      • save_video
      • process_video
  • VideoExtractionResult
    • Description
    • Attributes
    • Example Usage
  • VideoSegmentOutput
    • Description
    • Attributes
    • Example Usage
  • extract_videos
    • Description
    • Arguments
    • Returns
    • Example Usage
    • Saved Payload
    • Command-Line Interface
    • Related Module Utilities
  • TextMLPEncoder
    • Description
    • Parameters
    • Returns
    • Example Usage
  • Video3DEncoder
    • Description
    • Parameters
    • Returns
    • Example Usage
  • DynamicTarNetBase
    • Description
    • Parameters
    • forward
    • Example Usage
  • mse_loss
    • Description
    • Arguments
    • Returns
    • Example Usage
  • DynamicTarNet
    • Description
    • Parameters
    • Example Usage
    • Methods
  • DynamicGPIHyperparameterTuner
    • Description
    • Parameters
    • Search-Space Behavior
    • Example Usage
    • Methods
  • estimate_k_ipsi
    • Description
    • Data Arguments
    • Cross-Fitting Arguments
    • Dynamic TarNet Arguments
    • Downstream Nuisance-Model Arguments
    • Multimodal Encoder Arguments
    • Inference Argument
    • Returns
    • Example Usage
  • SpectralNormClassifier
    • Description
    • Parameters
    • Example Usage
    • Methods
      • forward
      • fit
      • predict_proba
      • predict
  • TarNet
    • Description
    • Parameters
    • Example Usage
    • Methods
      • create_dataloaders
      • fit
      • validate_step
      • predict
  • TarNetBase
    • Description
    • Parameters
    • forward
    • Example Usage
  • TarNetHyperparameterTuner
    • Description
    • Parameters
    • Example Usage
    • Methods
GPI: GenAI-Powered Inference
  • Search

© Copyright 2026, Kentaro Nakamura, Kosuke Imai.

Built with Sphinx using a theme provided by Read the Docs.